As a professional news and travel photographer, I’ve taken nearly one million photographs over my career to date.
I’ve photographed protests, rallies, concerts, births, homelessness, and udon soup, to name just a few subjects. I’ve taken photos on $5,000 mirrorless cameras from Leica, $2 film cameras from the 1940s, and everything in between.
All that’s to say that I know photos. So when the New York Times posed a provocative question in a recent article, I had strong feelings about it.
Their question? “Are AI images photographs?”
My answer? Absolutely not.
What Makes a Photo?
Photography requires two components, one technical and one conceptual.
The first is light. Fundamentally, a photograph is a moment of light, captured, fixed, and frozen forever.
The word “photography” literally means “drawing with light,” and as a photographer, you become exquisitely sensitive to it. I notice light wherever I go — the intensity, the color temperature, the angles.
Humans respond emotionally and intuitively to light, often without realizing it. A portrait taken outdoors under the warm, orange light of golden hour looks entirely different — and evokes different feelings in the viewer — than one taken under the harsh, blue light of a fluorescent-lit office.
Over photography’s 150+ year history, a lot has changed about how photographers capture and fix light. In the beginning, we relied on polished metal plates coated in toxic chemicals. For decades, we used tiny grains of silver sprayed onto long sheets of plastic. Today, we use light-sensitive digital pixels on a computer chip.
Still, though, the fundamentals haven’t changed. No matter what technical process we use, photography has always been about capturing light from the real world and preserving it indelibly.
The Act of Pointing
The other essential component of a photograph is more conceptual. The Times quotes curator John Szarkowski, who calls photography “the act of pointing.”
I love that definition.
When I teach photography, I have my students ask themselves what their photo is saying. If the answer is “Here’s a thing!” then the photo has fallen short.
Take food photography. Boring food photographs show a big plate of food sitting on a table. Everything is in focus. There’s nothing in the background. There’s no drama — no interest. The photo is saying “Here’s some noodles!”
A good food photo grabs something from the scene and points it out, amplifying it.
Maybe it’s a specific, important ingredient in the dish. Maybe it’s the environment in which the food is served — the decor of the restaurant, the items in the cook’s home. Maybe it’s a feeling — family, togetherness, tradition.
Beyond the craft of capturing light, photography requires noticing something in the real world and choosing to highlight it and pull it out from within a cacophony of competing things.
The art in photography comes from the photographer’s choice of what to pull out and amplify — what to point at. That “act of pointing” is how a photograph goes from simply documenting a thing, to expressing something essential that the photographer wants to convey about the world.
AI Images Aren’t Photographs
AI images lack both of these crucial aspects of photographs.
Firstly, AI images don’t use light. Unlike real photographs, the light in them is simulated — dreamed up the the silicon brain of a computer in a distant server farm.
Often, the creators of an AI-generated image has little control over the image’s lighting. DALL-E or Midjourney decide what kind of light to simulate based on the scene the tools are generating.
Ask for an image of a beautiful person, and you’ll probably get the soft light of a sunrise. Ask for a picture of a hacker, and you’ll get a person in a dark room (probably wearing a hoodie) bathed in the sickly red light of some unseen computer screen.
AI images’ pretend light is also frequently influenced by the source of images on which a given system was trained. That’s why today’s AI image generators struggle to create images that look average or mundane; they’re trained on billions of dramatic, colorful, highly saturated images found on Instagram or Pinterest, and so by default, they parrot back the “Instagram look.”
One of my favorite AI art projects is called Boring America Photorealism, by the artist John Kelly. Kelly used AI to re-create the kinds of mundane images we take on our cell phones every day.
Doing this with AI, it turns out, is incredibly hard. Kelly had to create hundreds of images and spend hours tweaking his prompts to get Midjourney, an AI image generator, to create a single, boring-looking result.
With real photography, the photographer’s end result is dictated by the light available to them in the scene. With AI-generated images, the image’s simulated light is dictated by its subject, the system’s training data, and the random hallucinations of a neural network.
Likewise, whereas real photography is about finding a subject and pulling it out of the infinite variety of the real world through “pointing,” AI images work differently.
AI typically creates the subject first and then fills in the details of the surrounding scene based on what feels like it should fit. It’s less about pointing at an object or concept in the real world, and more about choosing a subject and then creating a fake world that matches that subject.
Ask for a plate of latkes, for example, and you’re going to get a dreidel in the background. Ask for a board room, and you’re going to get white, male people in suits — unless the AI system’s creators have altered the results to inject artificial diversity.
Ultimately, the power of a photograph comes from the photographer’s real-world choices. The resonance of Napalm Girl comes both from the horror of the photo, and the fact that Nick Ut chose to capture and preserve that horrifying moment.
Photographs pull something from reality — a moment, a flash of light. AI-generated images do not.
The Value of AI Images
So does that mean that AI-generated images are artistically worthless?
Absolutely not. AI images aren’t photographs. But there’s a whole equally powerful branch of art that’s not directly derived from the real world, yet still has just as much artistic and explanatory power as photography: Illustration.
Illustrations are everything that photographs are not. Whereas photographs are constrained by reality — the laws of physics, the availability of subjects, the bounds of morality — illustrations are not. Illustrators are free to dream up any version of reality they’d like, and bring it into being with their pens or computers.
That’s part of why illustration is so powerful — and so dangerous. By imagining different political or cultural realities, satirical cartoonists challenge regimes and humble the powerful — it’s no surprise that historically, many have been assassinated.
Illustrations also have the power to help us imagine a potential reality that hasn’t yet come into being. Think about the illustrations in a patent application or the architectural renderings of a new skyscraper. Because these illustrations — unlike photos — aren’t constrained by the present reality, they can help us imagine a different future reality, which we can then actively create.
Like traditional cartoons or drawings, AI images are often powerful, resonant illustrations.
Elliott Higgins’ AI photos of Donald Trump’s imagined violent arrest resonate partly because they bring to life a fantasy of many liberals — that the former President would finally snap and reveal himself in an emperor-has-no-clothes moment that brings down his reign of power.
Making pictures of Trump getting arrested while waiting for Trump's arrest. pic.twitter.com/4D2QQfUpLZ
— Eliot Higgins (@EliotHiggins) March 20, 2023
Likewise, viral AI images of Pope Francis wearing a puffy Balenciaga jacket work because they reflect something important about Francis; with his progressive views on issues like gay marriage and women’s rights, we can imagine that Francis would do something as unusual (for the papacy) as wearing a puffer jacket.
AI images are fascinating because they take the traditional power of illustrations and amplify them, while also democratizing them.
Unlike most cartoons, AI images look like real photographs. It’s easier to suspend our disbelief when we see them, and therefore to slip gracefully into the imagined reality that the illustrator is choosing to show us.
And because tools like Midjourney and DALL-E create powerful illustrations in seconds and nearly for free, these tools open up the art of illustration to nearly anyone. My three-year-old can imagine a scene and use DALL-E to bring it to life.
Thanks to AI anyone with with an idea for a potential future reality can now illustrate it — no drawing skills required.
Again, AI images aren’t photographs. But as illustrations, they’re just as powerful.
No, they’re not constrained by the physical limitations of the real world. They’re not tied to light. They can’t point — or at least, they can’t point at reality.
They’re shaped by our conception of what certain subjects should look like, as expressed in the millions of photos we upload to Instagram and its ilk each day, and which AI gobbles up for training data.
But all those things are what makes AI images so revolutionary. Like the best illustrations, AI images help us dream up new versions of the world — or entire new worlds. They also reveal to us what we expect things to look like — an expectation we can accept or, like Kelly, challenge.
They reflect back to us our values, our fantasies (Trump-related or not), and our flaws.
In short, photographs are grounded in the real world. AI images transcend it.
I’ve tested thousands of ChatGPT prompts over the last year. As a full-time creator, there are a handful I come back to every day. I compiled them into a free guide, 7 Enormously Useful ChatGPT Prompts For Creators. Grab a copy today!
